What Research Interviews Can Teach a Friendship Startup—and What They Cannot

Conversations with potential members can be seductive. People describe loneliness, awkward invitations and the wish for a dependable circle, and a founder hears confirmation of an idea. But a compelling story is not the same as a tested behaviour. In building toward g3t, we have tried to take those conversations seriously while resisting the temptation to convert every agreement into a market verdict. This reflection is about how a team learns from people, not a claim that our research samples represent all adults.

The problem we chose to examine

A research interview can show the language a person uses to describe a frustrating experience. It may reveal that 'meeting people' and 'making friends' mean different things, or that the first hangout is less intimidating than asking somebody to meet again. It can also reveal ways we misunderstood our own explanation. Those are useful findings, especially when a team sees a pattern across several independently collected accounts. Yet the interview does not verify that the same person will pay, attend and return.

The design question underneath

Good questions make that distinction clearer. Instead of asking, 'Would you use a service that makes friendship easier?', ask about the last time someone met a new person, whether they suggested another meeting, who organized it and what actually happened. Concrete past behaviour has its own memory limitations, but it forces a more useful level of detail. Ask what was confusing and what felt unappealing as readily as what sounded promising. A polite yes should never close the research conversation.

What becomes clearer in real life

One lesson for a team is to keep different types of evidence separate. An interview quote is a perspective. A signup is an expression of intent. A payment is a transaction. Attendance is an event. Two people meeting a second time is another event. These markers can disagree. A promising application funnel might conceal weak repeat behavior; a small group with unremarkable signup figures might form stable friendships. Treating all those signals as one measure can lead to self-congratulation without learning.

What a good signal looks like

There is an ethical dimension too. People may share private stories about relocation, divorce or long periods without friends because the topic is personal. The dignity of those accounts matters more than a striking blog opening. We should not publish raw quotes, identities, transcript details or isolated percentages without a clear right to do so, an honest method and review. Synthesising themes without making participants identifiable is harder, but it is the standard a friendship company should seek.

Where our knowledge stops

Volunteer research samples tend to reflect people who can be reached, have time and are willing to talk. Prototype reactions are influenced by how the interviewer explains the concept. Participants may want to be kind to an earnest founder. Even rigorous qualitative analysis does not turn a small convenience sample into national prevalence. Research is therefore a tool for discovering questions and design risks, not a substitute for observing behaviour in the world. Our own previous conclusions should remain revisable.

The questions we need to keep testing

The useful next step after an interview is to identify the behaviour the product can test. If people report that plans fall apart in group chats, measure whether a clearer invitation increases actual attendance. If a second meeting is the stated need, observe real second meetings with consent rather than count only messages. Separate product quality from a person's private relationship outcomes. The research team should remain ready to hear the answer it least wants to hear.

Why the distinction matters

Interviews matter because they let us see a problem through another person's eyes. They become dangerous when we treat those eyes as a dashboard of universal truths. The point of asking is to build a better experiment, protect participants and keep learning after the prototype becomes a real evening in Paris. This first-person account requires founder review before publication.

The questions interviews cannot settle

An interview can uncover the words a person uses to describe an awkward second invitation. It cannot establish how often that problem occurs across an entire population unless the sampling design supports that conclusion. A person may tell us that they would pay for a thoughtfully arranged hangout and still decline when a real payment screen appears. A participant can like a prototype feature but never use it after launch. Those are not accusations of dishonesty; imagined decisions and real decisions happen under different constraints. We have to keep a boundary between reported preferences, observed behaviour and our own interpretation.

The strongest research loop therefore starts with an interview question, turns its answer into a testable product assumption, and measures a relevant outcome. Did members show up? Did they feel safe? Did they choose to meet another participant again, and did that next encounter actually happen? At every step we should count missing information rather than treating it as success. The people who never apply, withdraw or quietly disengage matter too. If our evidence mainly comes from the most enthusiastic participants, we risk designing for those easiest to satisfy rather than the adults who struggle most to find a dependable routine.